Contextual influences in the peripheral retina of patients with macular degeneration

Macular degeneration (MD) is the leading cause of low vision in the elderly population worldwide. In case of complete bilateral loss of central vision, MD patients start to show a preferred retinal region for fixation (PRL). Previous literature has reported functional changes that are connected with the emergence of the PRL. In this paper, we question whether the PRL undergoes a use-dependent cortical reorganization that alters the range of spatial lateral interactions between low-level filters. We asked whether there is a modulation of the excitatory/inhibitory lateral interactions or whether contextual influences are well accounted for by the same law that describes the integration response in normal viewers. In a group of 13 MD patients and 7 age-matched controls, we probed contextual influences by measuring the contrast threshold for a vertical target Gabor, flanked by two collinear high-contrast Gabors. Contextual influences of the collinear flankers were indicated by the changes in contrast threshold obtained at different target-to-flanker distances (λs) relative to the baseline orthogonal condition. Results showed that MDs had higher thresholds in the baseline condition and functional impairment in the identification tasks. Moreover, at the shortest λ, we found facilitatory rather than inhibitory contextual influence. No difference was found between the PRL and a symmetrical retinal position (non-PRL). By pulling together data from MD and controls we showed that in the periphery this inversion occurs when the target threshold approach the flankers’ contrast (about 1:3 ratio) and that for patients it does occur in both the PRL and a symmetrical retinal position (non-PRL). We conclude that contrary to previous interpretations, this modulation doesn’t seem to reflect use-dependent cortical reorganization but rather, it might result from a reduction of contrast gain for the target that promotes target-flankers grouping.

a two-dot configuration at the opposite side of the scotoma. This last result is compatible with cortical rewiring, whereby detection of the co-aligned low spatial filters crossing the scotoma becomes more efficient with MD. However, a more parsimonious explanation is that MD's peripheral vision takes its functional advantage from more efficient use of the high-level representation of the visual input. Indeed, the numerous functional changes that have been observed in MD vision 20,[24][25][26] are compatible with the suggestion that the visual response relies more on an integration field output.
To summarize, there is an important question that have no answer yet. Do the functional changes observed in the PRL result from modulation of excitatory and inhibitory contextual influences involved in contrast detection? To answer this question in this work we have looked at whether MD's peripheral vision is associated with a change in contextual influences with respect to controls and whether this change is accounted for by the same model that describes the contrast gain in normal viewer 27 . There are different ways in which the observed pattern of contextual influences in the periphery might be different in MD with respect to controls. In people with normal vision, accumulating psychophysical studies have shown contextual influences on the threshold for contrast detection coming from outside the receptive field of the channel responding to the target 1,[27][28][29] . In particular, it is well established that the visibility of a low contrast Gabor patch is affected by collinear flanking Gabors of similar orientation and spatial frequencies but high contrast: short target-to-flanker separation (1-2 times the wavelength of the target Gabor's carrier, λ) leads to suppression, whereas target-to-flanker separations of 3-4λ lead to enhancement [27][28][29][30][31][32][33] . Moreover, these contextual effects can be modulated by task repetition in a perceptual learning paradigm, in both normally sighted observers 33,34 and patients with impaired vision [35][36][37][38][39][40][41][42][43] . In addition, contextual influences depend on eccentricity: in the periphery, inhibition is more prominent and contextual enhancement occurs at a target-to-flanker distance of 6λ, that is larger than in the fovea 32,34,44 .
Since lateral connectivity has been shown to increase with practice in normal 31,34,45 and pathological vision [46][47][48] , use-dependent cortical reorganization might change the observed contextual influences. A partial support for the cortical reorganization hypothesis comes from recent papers showing that MD patients exhibited a modulation of contextual influences in respect of controls 35,49 . In the present study, we further address this issue attempting to specify the underlying mechanism. In addition, by comparing the effect of flankers distance on contrast detection both in the PRL and in a symmetrical retinal position (non-PRL) we also hope to address the still debated issue of whether the vision in the PRL is enhanced by the use of this region for everyday visual tasks 13,15,50,51 . A contextual modulations for target contrast detection with specific properties in the PRL would support the "Use-Dependent Reorganization" hypothesis while a strong similarity between the two tested locations would play in favour of a more conservative "Use-Independent Reorganization" hypothesis 13 . Finally, we asked whether a modulation of contextual influences in MDs affects the efficiency in performing everyday visual tasks.

Methods
Participants. Participants were 13 MD patients (mean age of 61 ± 9.16 years, range: 49-83 years) and 7 controls (mean age of 59 ± 4.12 years, range: 54-64 years). Patients were selected based on clinical history and Nidek MP1 microperimetry results. The dispersion of fixation was quantified during the microperimetry, and only patients with at least 80% of fixations in the range of 2° of visual angle around the focal point of the PRL were included in the sample. Patients with concomitant visual diseases other than central vision loss were not included, nor were those with a visual acuity on the ETDRS eye-chart lower than 1/20 or above 15/20. All the tests were performed monocularly and, although patients had a bilateral scotoma, the chosen eye was the one with the bestspared vision based on visual acuity and microperimetry data. The eye chosen to be tested for the control group was the non-dominant eye. Because the non-PRL was defined as the symmetrical retinal location of the PRL, the proximity of the optic disc or the irregular shape of the scotoma could reduce the visibility of stimuli presented in this second retinal location. To check for visibility of the stimuli, during both the crowding and lateral masking tasks, patients were asked to report the number of visible elements in the non-PRL. Of the 13 MD patients, all could see the full triplet of stimuli (Gabors and letters) in the PRL, while only 8 of them could do so in the non-PRL. Because the presentation of the stimuli was randomized in the two retinal positions to reduce the frequency and the amplitude of eventual eye movements, all the patients were tested in both locations, but only the ones who were able to see the full triplets were further considered for statistical analysis in the non-PRL position.
Details relative to age, gender, scotoma diameter, visual acuity, and PRL position are summarized in Table 1. The study was performed in accordance with the ethical standards laid down by the Declaration of Helsinki 52 . The study was approved by the Ethics Committee of the Department of General Psychology, University of Padova (Protocol 1449). We obtained written informed consent from all participants involved in the study.
Locations tested. The eccentricity of the PRL was individually estimated as a proportion of the distance from the macula and the optic disk, in degrees of visual angle. Using the image of the retinal fundus, the position of the fovea was computed based on the averages of the values determined for normally sighted observers: 15.3° temporally and 1.5° below the center of the optic disc. The distance between the position of the fovea and the PRL position estimated by the microperimetry was then computed to derive the eccentricity of the PRL. The non-PRL position was defined as approximately correspondent to the horizontally specular area of the retina, using the macula as the center of symmetry. One example of patients' microperimetry and the respective retinal displacement of the target Gabor in either the PRL or the non-PRL is shown in Fig. 1. In the case of very small scotoma, the non-PRL position was set by default at 6° of eccentricity with respect to the PRL on the horizontal axis. This was done to allow reliable discrimination between the two gaze positions by the eye tracker. The eccentricity at which stimuli were presented to each control subject matched that of one patient, randomly chosen. On average, the eccentricity was 4°.
www.nature.com/scientificreports www.nature.com/scientificreports/ Eye movement recording. Participants' fixation was controlled with an eye tracker to determine the retinal position corresponding to the patients' PRL/non-PRL and to be sure that fixation was maintained. Calibration and recording procedures were as follows. Eye movements were recorded using a Mirametrix S2 eye tracker with    www.nature.com/scientificreports www.nature.com/scientificreports/ a sample rate of 60 Hz and an accuracy of 0.5°. The calibration of the tracker and the gaze check were integrated into the main Matlab script using the Mirametrix Matlab Toolbox and API for Windows. Thanks to a custom calibration software, the calibration dot that normally sighted observers follow with the fovea was shifted by a constant so that, although MD patients could follow it with their PRL, the position of the eye relative to the calibration dot corresponded to that of a normally sighted observer. Due to the instability of fixation and the systematic error of the tracker, a tolerance window of ±1.5° around the PRL fixation point was set. If the gaze of the subject before the stimulus presentation was out of this window, a warning sound was presented to allow the patient to relocate his or her gaze.
Apparatus and stimuli. Participants sat in a dark room 57 cm from the screen. Stimuli were displayed on an ASUS ML228H LCD LED 21.5-inch monitor with a refresh rate of 60 Hz and a spatial resolution of 1920 × 1080 pixels, with a pixel pitch of 0.248 mm. Stimuli were generated with Matlab Psychtoolbox 53,54 . Gamma correction for each color channel was applied through calibration with the Spyder 4 Elite colorimeter (DataColor). The calibration was further verified using a Minolta LS-100 photometer, which indicated that the mean luminance was 50 cd/m2. In that way, luminance was a linear function of the digital representation of the image.
In order to represent 10.7 bits of luminance (1786 gray levels) on an 8-bit display, we adopted a software solution called "Pseudo-Gray, " also known as "Bit-Stealing" 55 , implemented via a Psychtoolbox built-in function.
Contrast detection stimuli. Stimuli were Gabor patches consisting of a cosinusoidal carrier enveloped by a stationary Gaussian. Each Gabor patch was characterized by its sinusoidal wavelength (λ), phase (ϕ), and standard deviation of the luminance Gaussian envelope (⌠) in the (x,y) space of the image: with ⌠ = λ and ϕ = 0 (even symmetric). Gabors' spatial frequency (SF) was 2 and 3 cycles/deg (cpd) for MD patients and 3 cpd for controls. A vertical low-contrast Gabor target (Fig. 2) was collinearly flanked, above and below, by two iso-oriented high-contrast Gabors (0.7 Michelson contrast). In addition, a condition with the vertical low-contrast Gabor target flanked by orthogonally oriented Gabors patches was added; with this stimulus configuration, the target detection is not modulated by lateral interactions 28,56 . The contrast threshold of the target was estimated according to a 1 up/3 down staircase. Participants performed a temporal two-alternative forced choice (2AFC). The target was presented in one of the two time intervals, whereas the flankers were always presented in both time intervals. Observers had to report in which time interval the target was presented. Feedback was provided for incorrect trials. Each block was terminated after 120 trials or 16 reversals. Contrast thresholds were estimated by averaging the contrast values corresponding to the last 8 reversals. For the PRL/non-PRL testing, contrast levels from two separated staircases were displayed in a random order over the two different retinal positions.
The two high-contrast collinear flankers were placed at various distances from the target (i.e., 2λ, 3λ, 4λ, and 8λ). The patients were asked to maintain their gaze on the PRL, and stimuli were randomly presented over either the PRL or non-PRL position within a block. Controls had to fixate the center of the screen, and stimuli were randomly presented either left or right of fixation.
Visual acuity and crowding stimuli. Visual acuity (eccentric VA) and crowding were measured at the same eccentricity as for the Gabor configuration. Stimuli were generated using Matlab Psychtoolbox 53,54 and presented at 57 cm. The stimuli were 10 letters (D, N, S, C, K, R, Z, H, O, and V) with Sloan 57 character type, randomly presented for 133 ms. The target letter was presented randomly at the two eccentricities in the same block, either the www.nature.com/scientificreports www.nature.com/scientificreports/ PRL/non-PRL for MD patients or 4° left/right from fixation for controls. The size of the letters for measuring acuity threshold and the edge-to-edge spacing for measuring crowding varied according to a psychophysical adaptive procedure (Maximum Likelihood Procedure) 58-60 that tracked 55% of the participants' psychometric function within a 60-trial block. The starting streak width was 30 arcmin. Subjects had to verbally report the target letter, and the experimenter registered the answer. The threshold was the values obtained in the last trial.
The crowding stimulus had two different letters vertically flanking the target letter. The streak width of both the target and flanking letters was set 30% higher than the VA threshold obtained at the same eccentricity and the same exposure duration. When tested in the PRL, the MD patients were able to identify all three letters at the largest spacing used (5°). This procedure is often used 34,36,61,62 to avoid an influence of VA on the measurement of critical spacing for crowding. Crowding was indexed by the critical spacing, defined as the edge-to-edge inter-letter distance at which observers could discriminate the target (the central letter) with 55% accuracy. Differently, from the center-to-center distance, edge-to-edge distance prevents overlay masking of the target by the flankers but has the disadvantage of co-varying with letter acuity (the bigger the target, the larger the center-to-center distance at zero border-to-border distance), thus ultimately underestimating crowding in people with low acuity.
Procedure. Each subject underwent a testing session of 3 hours in which VA, crowding, and the target contrast thresholds for orthogonal and several collinear configurations were measured monocularly in counterbalanced order.
Contextual influences were estimated by computing the threshold elevation (TE) as: 10 where CT_Collinear is the contrast threshold estimated in the collinear condition and CT_Orthogonal is the contrast threshold estimated in the orthogonal condition. TE was calculated separately for each target-to-flanker distance (i.e., 2λ, 3λ, 4λ, and 8λ). We used the orthogonal threshold value obtained at 8λ as a baseline to compute TE (instead of thresholds obtained from the isolated target). The advantage is that the orthogonal configuration still maintains the facilitation induced in the task by the reduced spatial and temporal uncertainty that the presence of the flankers causes so that it is not a confounding factor for the estimation of TE 63 .

Statistical analysis.
Within-and between-group comparisons were carried out with ANOVAs on either contrast threshold or TE using, for pairwise comparisons, t-tests with Bonferroni correction. TE was also analyzed using one-sample, one-tail t-tests based on the hypothesis of TE as either >0 or <0 for suppressive and facilitatory effects, respectively. Visual acuity and crowding data were also analyzed with Tukey tests.
Comparison of contrast thresholds and TE data obtained in the PRL and non-PRL of MD patients are shown in Fig. 4. Repeated-measures ANOVAs, including as factors the retinal locus (PRL and non-PRL) and the λ (2, 3, 4, and 8λ), were conducted on both the contrast thresholds and TE data of the subgroup of 8 patients that were tested with stimuli presented in both the PRL and non-PRL positions.
Acuity and crowding results. The visual acuity of the 8 patients who had a reliable measure in both the PRL and non-PRL are shown in Fig. 5. The one-way ANOVA conducted on these data, with group as a factor (controls, PRL, non-PRL) showed a significant effect of group (F(2,20) = 6.22, p = 0.007, partial-η2 = 0.384), indicating higher acuity for controls than MD patients both when tested at the PRL (difference = 6.917, p = 0.046) and non-PRL positions (difference = 9.167, p = 0.007). The difference between PRL and non-PRL was not significant (difference = −2.25, p = 0.67). Even considering the small sample and the high variability in VA data, this lack of difference confirms that development of a PRL is not strictly linked to an advantage in terms of visual acuity over the other retinal quadrants 64,65 . Moreover, the two acuity values for the 8 patients are highly correlated (R = 0.78, p = 0.022).
Individual crowding data of the 8 patients who had a reliable measure in both the PRL and non-PRL are shown in Fig. 6. The one-way ANOVA conducted on these data revealed a significant effect of group (F(2,20) = 3.516, p = 0.049, partial-η2 = 0.26). Post hoc comparison showed a significant difference between controls and MD patients when tested at the non-PRL (difference = 1.693, p = 0.039) but not when tested at the PRL (difference = 0.962, p = 0.3). The difference between PRL and non-PRL was not significant (difference = −0.731, p = 0.47). Visual acuity and crowding measures should be independent by definition. We checked this assumption by calculating the correlation between the two measures. We found that the negative correlation did not reach significance (R = −0.35, p = 0.18) despite the fact that the operative definition of critical spacing as the edge-to-edge inter-letter distance may have inflated this estimate (see Method section). www.nature.com/scientificreports www.nature.com/scientificreports/

Discussion
Using a two-interval forced choice task, contrast threshold for a low-contrast target Gabor flanked by two collinear high-contrast Gabors presented at eccentricities varying between 3° and 8° was measured in a group of subjects with MD and in an age-matched control group. Target-to-flanker separation, varied in terms of the Gabor's carrier wavelength unit (λ), was 2λ, 3λ, 4λ, and 8λ. The contextual influence of the flankers was quantified by the threshold modulation (TE), indicating the change in contrast threshold obtained at each of the four λs, relative to the baseline condition with no contextual influence (orthogonal flankers, 8λ).
Results showed contextual enhancement at λs higher than in the fovea (4-8λ). At 2λ, all controls had inhibition. Only 2 of the patients had inhibition, 2 had a TE close to zero and 9 of them had negative TE, indicating facilitatory contextual influences. This change in contextual influences at the shortest λ in MD patients was associated, both at the PRL and non-PRL, with an increase of contrast threshold for the target, as well as with reduced visual acuity and a larger crowding effect. Is the switch between inhibition and facilitation at the shortest λ an index of cortical plasticity or it could be explained by the same model used to interpret pshychophysical data from normal viewers? An answer to this question comes from establishing whether TE in MDs is well described by the variation of TE as a function of flanker/target contrast ratio in normal vision 31,66-69 . Zenger & Sagi (1996) proposed a model for contextual influences in which:  www.nature.com/scientificreports www.nature.com/scientificreports/ i. At a close distance, the sensitivity to a low contrast target is reduced by the presence of high contrast flankers. ii. When the contrast of target increases the reduction in sensitivity progressively decreases and then turns into facilitation. The switch happens when the contrast of the target is still lower (around three times) with respect to that of the flankers. iii. When the contrast of target approaches that of the flankers there is a deep in the facilitation. iv. However,when the contrast of target surpass that of the flankers the facilitatory effect progressively reduces and then disappear.
The model postulates a contrast dependent modulation of the contextual effect of the flankers that shift s progressively from inhibition to facilitation depending on flanker/target contrast ratio. We verified this by pulling the data of the two groups toghether and regressing the TE at 2λ as a function of log 10(Contrast flankers /Contrast target threshold ) in the 8λ orthogonal condition. We performed a locally-weighted polynomial regression (lowess) 70,71 in R 72 with a 50% smoothing span that leads to an R 2 of 0.52 (correlation between raw and estimated data). Finally, we superimposed raw data and fit line over the model predictions from Zenger and Sagi (1996) at 0λ and 2λ. As the Fig. 7 shows, the regression line derived form our dataset approximate very vell the one predicted by the model, in particular the line that refers to the 0 λ, as expected by the fact that increasing eccentricity would shift the curve leftwards. Thus, the transition from inhibition to facilitation that most MD patients show at high contrast threshold suggests improved efficiency in integrating/grouping elements, possibly mediated by an integration between the flanker and target within the 2 nd order integrative field 31 .
Our results do not support the hypothesis that MD present cortical reorganization leading to a use-dependent increase of long-range connectivity. In this case, with the low contrast target, we would have obtained increased contrast enhancement at the range of target-to-flanker distances at which facilitation occurs in normal vision. In the previous literature, the reduced collinear inhibition found in MD has been interpreted as a sign of neural plasticity, linked with a change in receptive field size 35,49 . We shed new light on this phenomenon and propose a different interpretation. If we consider together all our participants including the controls, our data show that the reduction in collinear inhibition and the switch towards facilitation are clearly linked with the baseline contrast sensitivity of the single subject in the orthogonal configuration. This switch can be well described by the same model previously proposed for the normal vision 31 and thus our results support the hypothesis that inhibitory target-flanker separations (short) become facilitatory in MD as it would do in controls if their contrast threshold were 4-5 fold higher. Taking all this information into consideration, the reduction of inhibition cannot be ascribed to neural plasticity in PRL but must be considered as a by-product of the same retinal degeneration that may deplete the patient's vision at the boundary of the scotoma. TEs as a function of log 10 (Contrast flankers /Contrast target threshold ) in the 8λ orthogonal condition are shown for the pooled data obtained by patients at the PRL (n = 13) and at the non-PRL (n = 8), and for the data of the control group (N = 7). Prediction for the 0 λ and 2λ based on the model from Zenger and Sagi (1996) are shown together with the locally weighted scatterplot smoothing (lowess) from our data.